Twin research has supported the concept of intelligence (general cognitive ability, g) by showing that genetic correlations between diverse tests of verbal and nonverbal cognitive abilities are greater than 0.50. That is, most of the genes that affect cognitive abilities are highly pleiotropic in the sense that genes that affect one cognitive ability affect all cognitive abilities. The impact of this finding may have been blunted because it depends on the validity of the twin method. Although the assumptions of the twin method have survived indirect tests, it is now possible to test findings from the twin method directly using DNA alone in samples of unrelated individuals, without the assumptions of the twin method. We applied this DNA method, implemented in a software package called Genome-wide Complex Trait Analysis (GCTA), to estimate genetic variance and covariance for two verbal tests and two nonverbal tests using 1.7 million DNA markers genotyped on 2500 unrelated children at age 12; 1900 children also had cognitive data and DNA at age 7. Because each of these individuals is one member of a twin pair, we were able to compare GCTA estimates directly to twin study estimates using the same measures in the same sample. At age 12, GCTA confirmed the results of twin research in showing substantial genetic covariance between verbal and nonverbal composites. The GCTA genetic correlation at age 12 was 1.0 (SE = 0.32), not significantly different from the twin study estimate of 0.60 (SE = 0.09). At age 7, the genetic correlations were 0.31 (SE =0.32) from GCTA and 0.71 (SE = 0.15).from twin analysis. The results from the larger sample and stronger measures at age 12 confirm the twin study results that the genetic architecture of intelligence is driven by pleiotropic effects on diverse cognitive abilities. However, the results at age 7 and the large standard errors of GCTA bivariate genetic correlations suggest the need for further research with larger samples.